posts

about this goblin thing, i made a video about! it was a lot of work, crazy. i had no idea how much work goes into putting yourself out there like this. but i did it! and i also wrote an article on my employer’s website, which i’m reposting below my video, enjoy! I was wondering how little power I could give an AI agent and still make it useful. I ended up writing a custom AI agent as part of a custom Kubernetes operator. I called it the Goblin Operator, and the agent being the Goblin Scout. You can find all the code here: github.com/sguldemond/goblin-operator ...

i was setting up IAM roles, permissions, service accounts etc for a project. in claude code i was testing custom permissions of two Service Accounts. one was only able to view the files in specific directory it uploaded itself, the other was able to view all files in that directory (not exactly, but basically). lets call them, self-viewer and all-viewer. when i had claude retrieve all files using the self-viewer SA, it told itself/me that “these are all the images in the directory, so the permission narrowing of self-viewer did not work”, i skipped over that remark because i could see in the output, where in fact there were only the 6 files the SA itself uploaded. ...

lets capture the moment we’re in, in a few years this is going to feel ancient again. i didn’t like the framework for the blog, mkdocs. mkdocs is fine, but doesn’t work well for a blog. not trying to do anything fancy, i do really like having my posts in markdown, so ended up going for hugo. still with the pre-ai thinking, “oh man, i have to redo the entire blog…” ...

could i create a small LLM agent with limited permissions, that can still be useful? yes i can, and i did! the idea is that when an issue occurs in the middle of the night, the goblin pings you on telegram (or slack, whatever). you see the proposed solution and you can chat with the goblin about it. OR you chat some more with it to come up with some fix, accept it and get back to sleep. OR you decide this needs deeper investigation, and get you a** out of bed. ...

oh man, i just wrote such a boring blog entry about my homelab. i mean, its not that boring per se, i like my homelab, but its just summing up all the technical aspects of it. an llm could do that better! what even could i tell about my homelab, without falling in the trap of boring myself again. well, working on it excites me. its not even doing much at the moment, its all 2012 era intel based mac hardware, so very underpowered. but thats the beauty of my field of engineering, i’m not per se limited by compute power. these machines run linux so well, everything is supported, they can even do some minimal virtualization. i have a vyos route creating a separate lan subnet, behind that a k3s cluster. i’m trying out all kinds of cool stuff. i’m serving a ignition/butane file over lan to when installing coreos on them, so i can super easily reinstall them. next i want to try out bootc. ...

inspired by lena dunham’s/hannah horvath’s quote: I think I might be the voice of my generation. Or, at least, a voice of a generation i’m starting a new blog, again. appareantly i tried already once before, i discovered when going through my github repos. but! new inspiration, new blog. but! why? well, because writing helps me think, and i like to think about things. i used to write a lot in my 20s, i was always carrying around a little note book, where i would write all kinds of stuff; practical notes, diary entries, drawings. recent years i’ve kinda only been writing documentation, not very existing, but helpful. ...